Rician Noise Removal in 3D MR Images Using Adaptive Non-Local Means Filter

Magnetic Resonance Images (MRIs) are contaminated with Rician distributed noise during acquisition process. Unlike Additive Gaussian distributed noise, noise in Magnetic Resonance Images follows the Rician distribution. Rician noise is signal dependent. In the high contrast region of the MR images, noise follows Gaussian distribution whereas in the low contrast regions it tends to be Rayleigh distribution. Therefore, separation of the Rician distributed noise in MR image signal is a challenging and difficult task. Non-Local Means (NLM) filter has been widely used to remove noise from the 2D natural images and 2D medical image sequences. In this paper, NLM filter is used for random noise filtering and adapted to remove the noise from MR image slices using redundancy of information in the image under study to remove the noise. Due to Rician nature of noise in MR images, noise reduction method is first applied to the squared magnitude of the images. The magnitude of MR images is the square ischolar_main of the sum of squares of Gaussian distributed real and imaginary parts of the image data that follows Rician distribution. Experimental results show that given methods achieves better denoising performance over the other existing noise removal methods.